opentui(phase3): launcher integration — HERMES_TUI_ENGINE dual-engine

hermes --tui launches the native OpenTUI engine (Bun) when
HERMES_TUI_ENGINE=opentui (env) or display.tui_engine=opentui (config);
Ink stays the default and the shipping path is untouched.

- _resolve_tui_engine() (env > config > ink); refuses opentui on
  Windows/Termux (no Bun) -> falls back to ink with a notice.
- _make_opentui_argv() -> [bun, src/entry.real.tsx] (no build step).
- _bun_bin() with HERMES_BUN override.
- Branch at top of _make_tui_argv BEFORE _ensure_tui_node (Bun-only host
  must not bootstrap Node).
- Gate _launch_tui NODE_OPTIONS/--max-old-space-size on engine==ink (Bun
  is JSC; the V8 flag errors/ignores).

Verified end-to-end via tmux: real hermes --tui -> Bun -> OpenTUI ->
real Python gateway streamed a real reply. No-flag default still ink.

run_agent.py

@@ -196,7 +196,7 @@ from agent.tool_dispatch_helpers import (
     _extract_error_preview,
     _trajectory_normalize_msg,  # noqa: F401  # re-exported for tests that `from run_agent import _trajectory_normalize_msg`
 )
-from utils import atomic_json_write, base_url_host_matches, base_url_hostname, is_truthy_value, model_forces_max_completion_tokens
+from utils import atomic_json_write, base_url_host_matches, base_url_hostname, is_truthy_value
 
 
 
@@ -358,7 +358,6 @@ class AIAgent:
         save_trajectories: bool = False,
         verbose_logging: bool = False,
         quiet_mode: bool = False,
-        tool_progress_mode: str = "all",
         ephemeral_system_prompt: str = None,
         log_prefix_chars: int = 100,
         log_prefix: str = "",
@@ -376,7 +375,6 @@ class AIAgent:
         thinking_callback: callable = None,
         reasoning_callback: callable = None,
         clarify_callback: callable = None,
-        read_terminal_callback: callable = None,
         step_callback: callable = None,
         stream_delta_callback: callable = None,
         interim_assistant_callback: callable = None,
@@ -432,7 +430,6 @@ class AIAgent:
             save_trajectories=save_trajectories,
             verbose_logging=verbose_logging,
             quiet_mode=quiet_mode,
-            tool_progress_mode=tool_progress_mode,
             ephemeral_system_prompt=ephemeral_system_prompt,
             log_prefix_chars=log_prefix_chars,
             log_prefix=log_prefix,
@@ -450,7 +447,6 @@ class AIAgent:
             thinking_callback=thinking_callback,
             reasoning_callback=reasoning_callback,
             clarify_callback=clarify_callback,
-            read_terminal_callback=read_terminal_callback,
             step_callback=step_callback,
             stream_delta_callback=stream_delta_callback,
             interim_assistant_callback=interim_assistant_callback,
@@ -1253,24 +1249,13 @@ class AIAgent:
     def _max_tokens_param(self, value: int) -> dict:
         """Return the correct max tokens kwarg for the current provider.
 
-        OpenAI's newer models (gpt-4o, gpt-4.1, gpt-5+, o-series) require
-        'max_completion_tokens'. Azure OpenAI and GitHub Copilot also require
-        'max_completion_tokens' for those families served via their
-        OpenAI-compatible endpoints. OpenRouter, local models, and older
+        OpenAI's newer models (gpt-4o, o-series, gpt-5+) require
+        'max_completion_tokens'. Azure OpenAI also requires
+        'max_completion_tokens' for gpt-5.x models served via the
+        OpenAI-compatible endpoint. OpenRouter, local models, and older
         OpenAI models use 'max_tokens'.
-
-        The check is URL-first (api.openai.com / Azure / Copilot all use the
-        new kwarg), then falls back to a model-name check so third-party
-        OpenAI-compatible endpoints fronting those models are recognised —
-        URL-only detection misses that case and silently sends the wrong
-        kwarg, which the upstream model rejects with a 400.
         """
-        if (
-            self._is_direct_openai_url()
-            or self._is_azure_openai_url()
-            or self._is_github_copilot_url()
-            or model_forces_max_completion_tokens(self.model)
-        ):
+        if self._is_direct_openai_url() or self._is_azure_openai_url() or self._is_github_copilot_url():
             return {"max_completion_tokens": value}
         return {"max_tokens": value}
 
@@ -2831,18 +2816,11 @@ class AIAgent:
         if state is None:
             return
         try:
-            from agent.credits_tracker import evaluate_credits_notices, is_free_tier_model
+            from agent.credits_tracker import evaluate_credits_notices
             latch = getattr(self, "_credits_latch", None)
             if latch is None:
                 latch = self._credits_latch = {"active": set(), "seen_below_90": False, "usage_band": None}
-            # Free-model gate: a depleted account on a free model can still
-            # inference, so the depleted error banner is suppressed. Local-data
-            # only (":free" suffix + pricing-cache peek) — never a network call.
-            model_is_free = is_free_tier_model(
-                getattr(self, "model", "") or "",
-                getattr(self, "base_url", "") or "",
-            )
-            to_show, to_clear = evaluate_credits_notices(state, latch, model_is_free=model_is_free)
+            to_show, to_clear = evaluate_credits_notices(state, latch)
             for key in to_clear:        # clears FIRST …
                 self._emit_notice_clear(key)
             for notice in to_show:      # … then shows (depleted lands last in a latest-wins slot)
@@ -3109,17 +3087,6 @@ class AIAgent:
         except Exception:
             pass
 
-        # 6. Free conversation history.  Mirrors _release_evicted_agent_soft's
-        # soft-eviction clear — close() is the hard teardown for true session
-        # boundaries (/new, /reset, session expiry), so the message list won't
-        # be reused.  Drops the reference proactively rather than waiting for
-        # the agent object itself to be collected, which matters when a caller
-        # still holds the closed agent (e.g. a draining background task).
-        try:
-            self._session_messages = []
-        except Exception:
-            pass
-
     def _hydrate_todo_store(self, history: List[Dict[str, Any]]) -> None:
         """
         Recover todo state from conversation history.
@@ -3937,13 +3904,6 @@ class AIAgent:
     def _anthropic_messages_create(self, api_kwargs: dict):
         if self.api_mode == "anthropic_messages":
             self._try_refresh_anthropic_client_credentials()
-        # Defensive: strip Responses-only kwargs that can leak in under an
-        # api_mode-flip race (the Anthropic SDK raises a non-retryable
-        # TypeError on them). See #31673.
-        from agent.anthropic_adapter import sanitize_anthropic_kwargs
-        sanitize_anthropic_kwargs(
-            api_kwargs, log_prefix=getattr(self, "log_prefix", "")
-        )
         return self._anthropic_client.messages.create(**api_kwargs)
 
     def _rebuild_anthropic_client(self) -> None: